Pointwise feature sensitivity analysis (Q1063395)

A general procedure to assess the quality and effectiveness of an assumed mathematical model studying the effect on some structural features of the output (critical points, integral transforms, etc.) by the change of the input parameters is suggested. The characteristics of interest of the mathematical model are represented by an ''operator equation'' of the form: \(F(O_{\alpha}| x)=0\) that may have several solutions, \(x^*_ j(\alpha)\), \(j=1,2,...,N\), giving the value of the variable x (representing certain features of the mathematical model) as functions depending on the input parameters \(\alpha =(\alpha_ 1,\alpha_ 2,...,\alpha_ k)\). The ''sensitivity coefficients'' are taken to be the partial derivatives \(\partial x^*_ j(\alpha_ 1,...,\alpha_ k)/\partial \alpha_ i\) and it seems that the largest part of the paper is used by the authors to derive the relations \(\partial x^*_ j(\alpha)/\partial \alpha_ i=(-\partial F(O_{\alpha}| x)/\partial \alpha_ i)/(\partial F(O_{\alpha}| x)/\partial x)|_{x=x^*_ j(\alpha)}\) giving the partial derivatives of the functions \(x^*_ j(.)\) implicitly defined by the above ''operator equation''. As an example, the results of the use of this procedure on the differential scattering cross section from a (12,6) Lennard-Jones potential are presented.